[app][macOS][26.707.91948] ChatGPT/Codex repeatedly grows to 55 GB and freezes the system
What version of the Codex App are you using (From “About Codex” dialog)?
26.707.91948 (build 5440)
What subscription do you have?
Unknown (signed-in Codex desktop user)
What platform is your computer?
Darwin 25.5.0 arm64 arm
- macOS 26.5.2 (25F84)
- MacBook Pro (MacBookPro18,3)
- Apple M1 Pro
- 16 GB RAM
What issue are you seeing?
The unified ChatGPT/Codex macOS app repeatedly grows until it exhausts application memory and freezes the whole computer.
The macOS Force Quit window reported:
- ChatGPT: 55.31 GB
- Claude: 5.69 GB
- Microsoft Teams: 3.66 GB
- Parallels Desktop: 4.30 GB
- Chrome: 897.4 MB
ChatGPT was by far the largest consumer. macOS displayed: Your system has run out of application memory. Multiple unrelated applications were marked as paused.
This is not a one-time event. Quitting or restarting ChatGPT/Codex temporarily restores memory usage to below approximately 1 GB, but the leak returns during continued Codex use and the machine freezes again.
The app was freshly auto-updated on 2026-07-16 and the installed build matches the latest official appcast entry, so there is no newer update available.
After restarting the app following the incident, macOS still showed approximately 2.07 GB of swap in use.
What steps can reproduce the bug?
- Open the unified ChatGPT desktop app and use Codex for local development tasks.
- Keep the app running through ongoing or long tasks.
- Memory usage grows over time.
- Eventually macOS reports that application memory is exhausted and pauses unrelated apps.
- Force quit or restart ChatGPT/Codex.
- Memory returns to normal temporarily, but the same problem recurs during subsequent use.
The workload includes code-mode tasks and may include MCP/browser workers. At the time of the post-restart inspection, the normal ChatGPT/Codex process tree included renderer processes, two node_repl processes, codex-code-mode-host, and the Codex app-server. No large orphan-process accumulation was visible after restart, so the exact leaking child from the incident could not be identified retroactively.
What is the expected behavior?
ChatGPT/Codex memory should stabilize at a reasonable level and reclaim renderer, browser, MCP, and Node worker resources after tasks complete. It must not consume tens of gigabytes, exhaust swap, or suspend unrelated applications.
The app should enforce an aggregate memory budget or stop and warn before causing a system-wide memory emergency.
Additional information
This reproduces on the latest build available on 2026-07-16 and appears related to:
- #33319
- #19333
- #18589
- #12491
The new evidence here is recurrence on version 26.707.91948 / build 5440 with the ChatGPT resource coalition reaching 55.31 GB on a 16 GB Apple Silicon Mac. A screenshot of the Force Quit window is available if maintainers need it.
This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗